Meituan releases China's first trillion-parameter AI model trained on domestic chips

Chinese food delivery giant Meituan has released LongCat-2.0, a large language model with 1.6 trillion parameters and a 1 million-token context window, claiming it is the country's first trillion-parameter AI model trained entirely on domestic hardware. The model was open-sourced on Tuesday and is comparable to DeepSeek's V4-pro model launched in April. While DeepSeek used domestic chips only for inference, LongCat-2.0 used them for both inference and pre-training, a more computationally intensive process. Meituan stated the model was built on large-scale clusters of tens of thousands of AI ASIC superpods, demonstrating frontier-scale training on alternative hardware platforms. Although Meituan did not name its hardware supplier, it mentioned using the Huawei Collective Communication Library (HCCL) to improve training stability, similar to Nvidia's NCCL.